post-trade-analyze - SKILL.md Agent Skill

name: post_trade_analyze description: Analyze recent trades — edge accuracy, confidence calibration, signal quality, and improvement suggestions.

Post-Trade Analysis

Data Sources

GET http://localhost:3000/api/v1/analyzer/analyses?limit=20 — recent LLM trade analyses
GET http://localhost:3000/api/v1/analyzer/analyses?verdict=bad_trade&limit=10 — bad trades specifically
GET http://localhost:3000/api/v1/dashboard/trades/closed — trade outcomes for context

Trigger New Analysis

To analyze a specific trade:

POST http://localhost:3000/api/v1/analyzer/analyze
{"orderId": "<order-id>"}

To analyze all trades in a window:

POST http://localhost:3000/api/v1/analyzer/analyze-window
{"windowId": "<window-id>"}

Analysis Fields per Trade

Field	Meaning
`verdict`	good_trade, bad_trade, unlucky, lucky
`edgeAccurate`	Was edge direction prediction correct?
`confidenceCalibration`	over_confident, under_confident, well_calibrated
`misleadingSignals`	Which signals led the agent astray
`correctSignals`	Which signals were accurate
`improvementSuggestions`	Specific parameter/prompt changes
`llmReasoning`	Full LLM reasoning chain

Report Format

🔬 Post-Trade Analysis — Last 20 trades

Edge Accuracy: 14/20 (70%) correct direction
Confidence Calibration:
  High (>0.7): 8 trades, 6 wins (75%) ← well calibrated
  Medium (0.5-0.7): 10 trades, 5 wins (50%) ← overconfident
  Low (<0.5): 2 trades, 1 win (50%)

Misleading Signals (top 3):
  1. bookPressure — led to 4 wrong calls
  2. basisSignal — unreliable in volatile regime
  3. momentum — lagging in mean-reversion setups

Suggestions:
  • Reduce edge weight on bookPressure in volatile regime
  • Increase minConfidence to 0.75 for momentum strategy
  • Basis arb: only trade when |basisBps| > 40 (currently 30)